✨ Core Features (Why GoChat?)

• 🔌 Unified Interface: Provides a consistent API interface that shields differences between different LLM providers
• 🏗️ Builder Pattern: Chain calls, complete LLM requests in a single line of code
• 🔐 OAuth2 Support: Built-in OAuth2 device code authentication for Qwen, Gemini, MiniMax
• 🔗 Workflow Orchestration: Elegantly organize complex RAG or Agent reasoning flows through Pipeline

📦 Installation

go get github.com/DotNetAge/gochat

🚀 Quick Start

1. Client - LLM Calls

Using the Builder pattern makes LLM calls extremely simple:

import "github.com/DotNetAge/gochat"

// Create Builder and send request
resp, err := gochat.NewClientBuilder().
    Init(gochat.Config{APIKey: "your-api-key"}).
    Model("gpt-4o").
    Temperature(0.7).
    UserMessage("Hello, please introduce the Go language").
    GetResponse(gochat.OpenAIClient)

if err != nil {
    log.Fatal(err)
}
fmt.Println(resp.Content)

Streaming Response:

stream, err := gochat.NewClientBuilder().
    Init(gochat.Config{APIKey: "your-api-key"}).
    Model("gpt-4o").
    UserMessage("Write a poem about spring").
    GetStream(gochat.OpenAIClient)

if err != nil {
    log.Fatal(err)
}
defer stream.Close()

for stream.Next() {
    event := stream.Event()
    fmt.Print(event.Content)
}

Supported Client Types:

More Builder Methods:

gochat.NewClientBuilder().
    Init(config).
    Model("gpt-4o").           // Set model
    Temperature(0.7).          // Set temperature
    MaxTokens(1000).           // Max tokens
    TopP(0.9).                 // Top-p sampling
    Stop("###", "END").        // Stop sequences
    EnableThinking(true).      // Enable thinking mode
    ThinkingBudget(1024).      // Thinking budget
    EnableSearch(true).        // Enable search
    SysemMessage("You are an assistant"). // System message
    UserMessage("User message").   // User message
    AssistantMessage("Assistant message"). // Assistant message
    AttachFile(attachment).    // Attach file
    AttachImage(image).        // Attach image
    Tools(tool).               // Tools
    UsageCallback(func(u core.Usage) {
        fmt.Printf("Token usage: %d\n", u.TotalTokens)
    }).
    GetResponse(gochat.OpenAIClient)

2. Auth - OAuth2 Authentication

GoChat has built-in OAuth2 device code authentication support for Qwen, Gemini, and MiniMax:

Qwen

import "github.com/DotNetAge/gochat/auth"

// Create AuthManager
manager := auth.Qwen()

// Login to get token (will print verification link and code)
if err := manager.Login(); err != nil {
    log.Fatal(err)
}

// Get token for API calls
token, err := manager.GetToken()
if err != nil {
    log.Fatal(err)
}

// Use token to create client
resp, err := gochat.NewClientBuilder().
    Init(core.Config{AuthToken: token.Access}).
    Model("qwen-max").
    UserMessage("Hello").
    GetResponse(gochat.OpenAIClient)

Gemini

// Need to provide OAuth2 configuration
manager := auth.GeminiWithConfig(
    "your-client-id",
    "your-client-secret",
    "http://localhost:8080/callback",
    ":8080",
)

if err := manager.Login(); err != nil {
    log.Fatal(err)
}

MiniMax

// "cn" for China version, other values for international version
manager := auth.MiniMax("cn")

if err := manager.Login(); err != nil {
    log.Fatal(err)
}

Token Persistence:

// Specify token file path
manager := auth.Qwen("/path/to/qwen_token.json")

// First login
manager.Login()

// Automatically load saved token on subsequent startups
token, err := manager.GetToken()

3. Pipeline - Workflow Orchestration

Pipeline is used to organize complex LLM workflows:

import (
    "github.com/DotNetAge/gochat"
    "github.com/DotNetAge/gochat/pipeline"
    "github.com/DotNetAge/gochat/pipeline/steps"
)

// Create client
client, _ := openai.NewOpenAI(core.Config{
    APIKey: "your-api-key",
    Model:  "gpt-4o",
})

// Create Pipeline
p := pipeline.New[*pipeline.State]().
    AddStep(steps.NewTemplateStep(
        "Please answer the following question: {{.question}}",
        "prompt",
        "question",
    )).
    AddStep(steps.NewGenerateCompletionStep(client, "prompt", "answer", "gpt-4o"))

// Create state and set input
state := pipeline.NewState()
state.Set("question", "What is Go language?")

// Execute Pipeline
if err := p.Execute(ctx, state); err != nil {
    log.Fatal(err)
}

// Get result
fmt.Println(state.GetString("answer"))

Built-in Step Types:

Add Hook to Monitor Execution:

type LoggerHook struct{}

func (h *LoggerHook) OnStepStart(ctx context.Context, step pipeline.Step[*pipeline.State], state *pipeline.State) {
    fmt.Printf("[%s] Started\n", step.Name())
}

func (h *LoggerHook) OnStepComplete(ctx context.Context, step pipeline.Step[*pipeline.State], state *pipeline.State) {
    fmt.Printf("[%s] Completed\n", step.Name())
}

func (h *LoggerHook) OnStepError(ctx context.Context, step pipeline.Step[*pipeline.State], state *pipeline.State, err error) {
    fmt.Printf("[%s] Failed: %v\n", step.Name(), err)
}

p := pipeline.New[*pipeline.State]().
    AddStep(step1).
    AddStep(step2).
    AddHook(&LoggerHook{})

🔌 Fully Supported Providers

🏗️ Project Architecture

gochat/
├── gochat.go       # Builder pattern entry
├── client/         # LLM provider client implementations
├── core/           # Core interfaces and common functionality
├── auth/           # OAuth2 authentication module
├── pipeline/       # Workflow orchestration functionality
├── provider/       # Additional provider implementations
├── docs/           # Documentation
└── examples/       # Example code

📄 License

This project is open-sourced under the MIT License. PRs are welcome!

📚 Comprehensive Documentation

Please refer to the docs/ directory for detailed guides, architecture diagrams, and API references:

• 📖 Project Overview
• 🚀 Quick Start
• 🧠 Client Module (LLM & Tool Calling)
• ⛓️ Pipeline Module (Workflow Orchestration)
• 🏢 Provider Module (OAuth2 Authentication)
• 📋 API Reference